How successful your next R package will be? modeling tool

This is a website of the project “How successful your next R package will be? A prediction model using R packages features” by:

1. Daniel Antiporta,
2. Marta Karas,
3. Perry Kuo.

The website is a part of the final assignment in 140.712.01 Advanced Data Science II class of 2018/19 at JHSPH.

Location of project’s screencast

Two minute screencast with narration showing a demo of the project can be accessed by clicking the image below (equivalently, click here).

Location of project’s GitHub repository

The project’s GitHub repository can be found here: https://martakarass.github.io/712-final_project/

In particular, the project’s GitHub repository contains:

• final RMarkdown file: 2018-12-09-project-summary.Rmd (link)

• corresponding compiled HTML file: 2018-12-09-project-summary.html (link)

Motivation and Overview

How many times you have found yourself spending long, long hours wrapping up an R package, polishing and pushing it to CRAN, and realize after that almost nobody downloads it and uses it?

Hence, you may keep asking yourself:

• How successful my next R package will be?
• What makes it successful?

Here we come, with the How successful your next R package will be? modeling tool that analyzes your package prototype based on its:

1. title and description text,
2. meta data,
3. code files content,
4. attached data content,
5. vignettes content

and predict a number of downloads it will generate over the time!

Project objectives

1. Develop a predictive model that takes as an input package’s features and predicts a number of downloads it will generate over (a) 3 months, (b) 1 year.

2. Identify what features of an R package derived from package’s features such as: title and description text, metadata, code files content, attached data content, vignettes content etc. are associated with a high number of downloads.

Data sources: examples

Data sources we used include:

• R packages’ description sites (from CRAN) information,
• R packages’ archive’s files (from CRAN) to access information about the package from its 1st release version,
• Utilizing already existing tools to get selected package’s statistics (metadata, download numbers etc.)

Exemplary data source: R package CRAN site

Derived explanatory variables: text mining

Examples of words (word’s cores) in features generated from package’s title (LEFT HAND SIDE) and package’s description (RIGHT HAND SIDE):

Derived explanatory variables: examples

Examples of words (word’s cores) in features generated from package’s meta data, code files, attached data content, vignettes:

Utilized modeling approaches

We have trained and tuned three different types of predictive model:

Results: prediction MSE on test set

The prediction results of the outcome logarithm of number of downloads over 1 year time on the test set are shown below.

Results: explanatory variables

The top variables in terms of:

• for linear regression: absolute value of coefficient estimate (note data frame of explanatory variables has been standardized to have mean 0 and variance 1),
• for random forest: variable importance statistic

from modeling the outcome logarithm of number of downloads over 1 year time are shown below.

Conclusion

• Random Forest and Support Vector Machine minimize the most MSE in our test dataset.

• The top 5 variables (variable importance rank) capture features related to packages: title (“ggplot”), authors (number), unit testing (files), description (“interface”), data files.

• R package developers should include the top features highlighted in our results to increase the ‘success’ of their products.